ABOUT THE TEAM

ARTAX is an experimental project focusing on the Machine Learning supported creative process in the context of music. The project started with a simple idea to look at how AI can enable greater freedom for the artist during the producing or composing process, creating new sounds or complementing the piece’s composition. The team was formed in late 2020 by IAMYANK, a composer and electronic music producer responsible for the artistic vision, and Adam Süki, a music enthusiast data scientist who supports the efforts from a technological point of view.

ABOUT THE SONG

Our approach was to put the creator in focus and look at AI as a tool that extends the creative personality. We aimed to create a song that is coherent with the artist’s current algorithmic composing aspiration. We also wanted the song to feel like his own piece with a twist. The surprising added value came from the concept of imprinting the composer's musicality within the Machine Learning model. The song’s first notes were created by a machine trained on adolescent inspirations such as Radiohead, Muse, Nirvana, and Deftones.

ABOUT THE HUMAN-AI CO-CREATION PROCESS

Before we started the compositional process, we had several trial and error experiments to determine how we will work: Which tools are we allowed to use, what limitations we set to ourselves, where can a human interact with the process, and what should be generated?

The central musical motifs of the song are based on parts of lyrics generated by a neural network that taught to write poetry based on inspirational bands’ work. After the texts were curated, some of its phonetic elements were assigned to pitches, thereby giving structure to the song.

Other voices of the song can be traced back to the generated main melody. The pitches created from the lyrics were stretched, offsetted, and transposed based on simple mathematical rules to evoke a form of composition based on classical, constantly evolving motifs.

We have experimented with other existing machine learning tools, some of which were used to complement the final piece. Neural networks provided by the Magenta research project were used to create drum patterns. We generated the atonal atmospheric sounds with NSynth, a neural synthesizer also by the Magenta project.

While we are sure that the final piece would not sound so unified without our artist’s take on the concept and personal touch on the productional qualities, we appreciate that we have managed to achieve a creative workflow that feels natural within this novel framework.